Optimized Backstepping-Based Containment Control for Multiagent Systems With Deferred Constraints Using a Universal...

Area of Science:

• Control Systems Engineering
• Artificial Intelligence
• Robotics

Background:

• Multiagent systems (MASs) present challenges in coordinated control, especially with state constraints.
• Deferred full-state constraints require advanced control strategies to ensure system stability and performance.
• Existing methods often struggle with initial condition restrictions and dynamic constraint handling.

Purpose of the Study:

• To develop an optimized containment control strategy for MASs with deferred full-state constraints.
• To propose a universal nonlinear transformation for handling both constrained and unconstrained scenarios.
• To ensure follower states converge within the convex hull of leader states.

Main Methods:

• A universal nonlinear transformation is employed to manage state constraints flexibly.
• A state-shifting function is utilized to eliminate initial restriction conditions.
• A neural network (NN)-based reinforcement learning (RL) algorithm with an identifier-critic-actor architecture is implemented.
• The Hamilton-Jacobi-Bellman (HJB) equation is integrated into subsystems for performance optimization.
• Simplified NN updating laws based on gradient descent are used.

Main Results:

• The proposed method effectively handles deferred full-state constraints by forcing states back within boundaries within a set time.
• Initial state restrictions are overcome by shifting states to the midpoint of boundaries.
• Lyapunov stability theorem and graph theory confirm signal boundedness.
• Simulations demonstrate that follower states converge to the convex hull formed by leader states.

Conclusions:

• The developed approach provides a robust and effective solution for optimized containment control in MASs with deferred full-state constraints.
• The NN-based RL strategy ensures stability and convergence, validating the proposed methodology through simulations.